A method, system, device and storage medium for detecting endometrial cancer cells

Abstract

The present invention discloses a method, system, device and storage medium for detecting endometrial cancer cells, including: preparing endometrial cell pathological sections; preliminarily screening images containing positive cells by using CNN; performing intelligent detection of cancer cells by using an improved Faster R-CNN object detection model; counting various indicators, generating patches from WSI and sending them into the network for prediction, and then merging the recognition results. In the part of target acquisition, the data set is normalized and expanded by using the method of adaptive random cropping; on the model, an improved CNN model and a two-stage Faster RCNN model are used as the basic models, and the doctor's diagnosis experience is introduced, and diagnostic indicators are extracted and jointly used to train a more accurate model. The accuracy of the endometrial cancer cell detection method of the present invention is higher than that of the classification method, and at the same time, the confidence level and the position information of the WSI can be given. In addition, the present invention also provides a more effective method for preparing an endometrial cell pathological section image data set. The present invention is applied to the field of medical image object detection.

Description

Technical Field

[0001] The present invention belongs to the technical field of AI medical imaging, and relates to a method, system, device and storage medium for detecting endometrial cancer cells. Background Art

[0002] Endometrial cancer is a group of epithelial malignant tumors that occur in the endometrium and is one of the three major malignant tumors of the female reproductive system. There are more than 378,000 patients worldwide every year, and its incidence rate is on the rise worldwide and tends to be younger. In addition, research shows that the 5-year survival rate of early endometrial cancer after appropriate treatment exceeds 85%. Early screening and early diagnosis of precancerous lesions of endometrial cancer can greatly reduce its incidence and mortality.

[0003] Computer-aided diagnosis systems can help doctors effectively analyze and evaluate a large number of medical images. It has been applied to the radiological diagnosis of some common cancers such as breast cancer, lung cancer and colon cancer, showing the expert level of human disease classification. In recent years, with the development of artificial intelligence technology and the improvement of computer computing power, the application of deep learning in medical analysis is considered to be the third eye of doctors. Deep learning algorithms based on deep convolutional neural networks have been proven to be a powerful driving force for advancing biomedical image analysis. Medical image computer-aided diagnosis systems based on deep learning are also increasingly being introduced into the clinical diagnosis work of hospitals, which can reduce the workload of pathologists and provide decision-making assistance. Therefore, how to use the rich data of clinical images and artificial intelligence methods to assist in the diagnosis of endometrial cancer has become the focus of attention of researchers.

[0004] High-tech modern means based on artificial intelligence can provide guarantee for the healthy life of the vast number of social women. It enables major malignant diseases such as endometrial cancer to be detected, diagnosed and treated early, reducing the pain and economic pressure of the vast number of women patients, and having very important social benefits. Proposing intelligent, simple and economical endometrial cancer screening methods can more effectively prevent cancer, detect it in time and take treatment strategies in time. If we can apply existing knowledge and technologies more widely and focus on high-risk groups, it will surely accelerate our progress in the fight against cancer.

[0005] Through the research results of computer-aided diagnosis systems in recent years, it can be found that artificial intelligence has played an excellent role in many cancer screenings. However, due to the characteristics of difficult cancer diagnosis, cell overlap, adhesiveness, etc. in the endometrium, at present, more reliance is placed on pathologists for manual screening at home and abroad. The number of pathologists is small, and excellent pathological talents are mostly concentrated in hospitals in big cities. The level of pathologists in grass-roots hospitals varies, and the lack of pathological talents further aggravates the burden on physicians and increases the probability of misdiagnosis and missed diagnosis. Artificial intelligence can effectively assist pathologists in screening pathological pictures, greatly reducing the workload of pathologists and being widely used in clinical diagnosis. At present, the applications of artificial intelligence in pathological diagnosis mainly include primary screening of tumor cytology, qualitative and quantitative analysis. In histopathological diagnosis, it is mainly used to assist in tumor prognosis judgment, histological classification, and benign and malignant differentiation, and certain progress has been made in lung cancer, prostate cancer, and cervical cancer. Research shows that artificial intelligence will provide reliable computer-aided diagnosis decisions for diagnostic pathology, and its robustness and objectivity will greatly improve the working environment of pathologists and bring effective economic benefits to society. Summary of the Invention

[0006] The purpose of the present invention is to solve the problems of slow time consumption, low accuracy, and strong subjectivity in the diagnosis process of existing endometrial cancer cell pathological images, and to provide a method, system, device, and storage medium for detecting endometrial cancer cells.

[0007] To achieve the above object, the present invention adopts the following technical solutions:

[0008] A method for detecting endometrial cancer cells includes the following steps:

[0009] Use a CNN model to perform primary screening on the WSI image of the endometrial cell pathological specimen to screen out the image containing positive cell samples;

[0010] Use the Faster R-CNN object detection model to identify the image of the positive cell sample;

[0011] Restore the WSI image corresponding to the identified endometrial cancer cells to obtain the position and confidence of the endometrial cancer cells.

[0012] The further improvement of the above method is as follows:

[0013] The WSI image is collected from the endometrial cell specimen by a Li-brush endometrial cell collector and scanned by a digital scanner to obtain the WSI image.

[0014] The use of the CNN model to perform primary screening on the WSI image of the endometrial cell pathological specimen to screen out the image containing positive cell samples includes:

[0015] First, the WSI image is segmented using the openslide tool. A complete cytopathological WSI image is input, and several 1024×1024 image patches are output. The image patches are input into the CNN network for classification screening, and the preliminary screening results are output.

[0016] The CNN network is a CNN model based on Attention and Bag of words. The CNN model uses the VGG network as the backbone network, and channel attention and spatial attention modules are added to both sides of the backbone network. Finally, the feature maps of the three channels are fused as the extracted features. Then, the high-dimensional features are extracted for k-means mean clustering, and the clustering centers are used as the codebook. Then, the statistical vectors of the codebook are used as the features of the samples, and finally, they are input into the SVM classifier for classification. The final model realizes the preliminary screening classification of positive and negative samples.

[0017] Using the Faster R-CNN object detection model to identify the images of positive cell samples includes:

[0018] Processing is carried out using the method of online random cropping. Randomly select a target in the input image of the positive cell sample, randomly cut out a sub-image with a side length in the range of 256 to 512 pixels around the target, and then scale it to a side length of 512 pixels and input it into the Faster R-CNN object detection model for identification. If the side length of the target exceeds 512 pixels, the target and the background are directly cut out and then scaled.

[0019] Restoring the WSI image corresponding to the identified endometrial cancer cells to obtain the position and confidence of the endometrial cancer cells includes:

[0020] Restore the image of the positive cell sample to 1024×1024 pixels, and completely restore the WSI image of the endometrial cell pathological specimen according to the row number and column number. The WSI image contains the position and confidence of the cancer cells identified by the Faster R-CNN object detection model.

[0021] An endometrial cancer cell detection system includes:

[0022] An image acquisition module for using the CNN model to perform preliminary screening on the WSI image of the endometrial cell pathological specimen to screen out the images containing positive cell samples;

[0023] An identification module for using the Faster R-CNN object detection model to identify the images of positive cell samples;

[0024] A position and confidence calculation module, which is used to restore the WSI image corresponding to the identified endometrial cancer cells to obtain the position and confidence of the endometrial cancer cells.

[0025] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above method are implemented.

[0026] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention proposes an intelligent, minimally invasive, and low-cost endometrial cancer screening method based on artificial intelligence, which is expected to liberate pathologists from the cumbersome work of reading endometrial slices, is expected to reduce the huge economic burden on families caused by the disease, and at the same time reduce the loss of labor caused by the disease and reduce social and economic losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0030] Figure 1 It is a flowchart of the detection method of the present invention.

[0031] Figure 2 It is a schematic diagram of the principle of the detection system of the present invention.

[0032] Figure 3 It is a schematic diagram of the sampling process of the endometrial sampling specimen provided by the present invention.

[0033] Figure 4 It is a schematic diagram of the process of making an endometrial cell pathological section image provided by the present invention.

[0034] Figure 5 It is the initial screening CNN network model of the present invention.

[0035] Figure 6 It is the endometrial cancer cell recognition network model of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0036] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. Components of the embodiments of the present invention usually described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0037] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0038] It should be noted that like reference numerals and letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0039] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper", "lower", "horizontal", "inner", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the drawings, or the orientations or positional relationships in which the inventive product is customarily placed during use, it is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation of the present invention. In addition, terms such as "first", "second", etc. are only used for descriptive distinction and cannot be construed as indicating or implying relative importance.

[0040] In addition, if the term "horizontal" appears, it does not mean that the component is required to be absolutely horizontal, but it can be slightly inclined. For example, "horizontal" only means that its direction is more horizontal relative to "vertical", and does not mean that the structure must be completely horizontal, but it can be slightly inclined.

[0041] In the description of the embodiments of the present invention, it should also be noted that unless otherwise clearly specified and limited, if terms such as "set", "installed", "connected", "connected" are understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0042] The present invention will be further described in detail below with reference to the accompanying drawings:

[0043] SeeFigure 1 , an embodiment of the present invention discloses a method for detecting endometrial cancer cells, including the following steps:

[0044] Step S1, using a CNN model to perform a preliminary screening on the WSI image of an endometrial cell pathological specimen to screen out the images containing positive cell samples; the WSI image is obtained by automatically collecting an endometrial cell specimen using a Li-brush endometrial cell collector and scanning it with a digital scanner to obtain the WSI image. The WSI image is segmented using the openslide tool. Input a complete cell pathological WSI image, and output several image patches of 1024×1024 pixels; input the image patches into the CNN network for classification screening, and output the results of the preliminary screening. The CNN network is a CNN model based on Attention and Bag of words; the CNN model uses the VGG network as the backbone network, and channel attention and spatial attention modules are added to both sides of the backbone network respectively. Finally, the feature maps of the three channels are fused as the extracted features, and then the high-dimensional features are extracted for k-means mean clustering. The clustering centers are used as the codebooks, and then the statistical vectors of the codebooks are used as the features of the samples. Finally, the samples are input into the SVM classifier for classification, and the final model realizes the preliminary screening classification of positive and negative samples.

[0045] Step S2, using the Faster R-CNN object detection model to identify the images of positive cell samples; using the method of online random cropping for processing; randomly select a target in the input image of the positive cell sample, randomly cut out a sub-image with a side length in the range of 256 to 512 pixels around the target, and then scale it to a side length of 512 pixels and input it into the Faster R-CNN object detection model for identification; if the side length of the target exceeds 512 pixels, then directly cut out the target and the background, and then perform scaling.

[0046] Step S3, restore the WSI image corresponding to the identified endometrial cancer cells to obtain the position and confidence of the endometrial cancer cells. Restore the image of the positive cell sample to 1024×1024 pixels, and completely restore the WSI image of the endometrial cell pathological specimen according to the row number and column number. The WSI image contains the position and confidence of the cancer cells identified by the Faster R-CNN object detection model.

[0047] As Figure 2 shown, an embodiment of the present invention also discloses an endometrial cancer cell detection system, including:

[0048] An image acquisition module, configured to use a CNN model to perform a preliminary screening on the WSI image of an endometrial cell pathological specimen to screen out the images containing positive cell samples;

[0049] An identification module for identifying an image of a positive cell sample using a Faster R-CNN object detection model;

[0050] A position and confidence calculation module for restoring the WSI image corresponding to the identified endometrial cancer cells to obtain the position and confidence of the endometrial cancer cells.

[0051] Example:

[0052] The intelligent detection method for endometrial cancer cells based on Faster R-CNN of the present invention has the following specific identification steps:

[0053] Step 1: Clinically collect the endometrial cell pathological specimens of the patient to be examined through a Li-brush endometrial cell collector, and scan the specimens through a special scanner to make a WSI containing hundreds of millions of pixels.

[0054] First, the user brushes the endometrial cell biological specimen of the patient with a Li-brush, and then places the prepared glass slide with the cell sample on the stage for scanning to obtain the full pathological image of the patient's endometrial cells.

[0055] This step is the basis for the network training of the subsequent steps. A specimen with high clarity and uniform staining is of great significance for computer recognition. Therefore, we have developed a standardized and unified acquisition and slide-making process:

[0056] Sample collection process:

[0057] ① The patient lies on the examination bed in the lithotomy position, is routinely disinfected and covered with a drape. With the help of a disposable cervical clamp and probe, the endometrial cell collector is inserted into the uterine cavity, and the brush hairs are rotated to collect the endometrial sample.

[0058] ② Place the sampling brush after sampling into the endometrial cell preservation solution, and swing the brush head to make the endometrial sample fall off into the preservation solution for preservation.

[0059] Slide-making and cell block embedding process:

[0060] ① After centrifugally enriching the sample in the preservation solution, add it to the slide-making machine and centrifuge again, fix and stain with 95% alcohol to obtain a liquid-based cell smear.

[0061] ② Add the enriched sample to the diluent, quickly put it into the embedding machine. After the sample diluent solidifies, take out the solid specimen diluent, cut off the part without the cell layer, and put it into the embedding cassette for embedding.

[0062] ③ Section the embedded cell block in the pathology department to obtain cell and microtissue sections.

[0063] The production of the medical pathology dataset involves the personal privacy of patients. The patients were recruited from the First Affiliated Hospital of Xi'an Jiaotong University from July 2015 to July 2019. The cytopathological sections were collected from women who underwent curettage or hysterectomy. This study was approved by the Ethics Committee of the First Affiliated Hospital of Xi'an Jiaotong University, and all patients signed written informed consent forms. All protocols used complied with the ethical principles regarding human subjects in the Declaration of Helsinki of Medical Research.

[0064] Patients diagnosed with suspected pregnancy or pregnancy, acute reproductive system inflammation, cervical cancer, or coagulation disorder diseases were excluded. Women with a body temperature of 37.5°C or above measured twice a day were also excluded. The following table is the statistical information of the medical characteristics of the collected patients.

[0065] Table 1 Statistical Table of Patient Characteristic Information

[0066]

[0067] Step 2: Use CNN to preliminarily screen the images containing positive cells.

[0068] Read the whole cytopathological image prepared by the patient in Step 1. At the same time, call the system segmentation tool openslide to segment the WSI. When segmenting, ensure that the naming format of each sub-image is patient number_row number_column number.jpg, and save them in the same specified folder. Read the trained CNN model with an attention mechanism, put all the sub-images in the folder into the queue of images to be segmented. Each batch input into the CNN model is 32 sub-images. It should be understood that with the change of machine performance, the size of the batch can be increased or decreased accordingly. Output all sub-images with a positive scoring rate greater than the threshold by the network, save these sub-images in Folder 1, and save the remaining sub-images in Folder 2 for future restoration of the WSI. Here, the instance threshold is selected as 0.8, and the threshold can be changed according to needs. It should be noted that when training this CNN model, accurate labels need to be given by professional pathologists. Generally, two professional pathologists with rich clinical experience and long working years are selected to give labels to the dataset. Pathologist A gives the labels, and Pathologist B reviews the results given by Pathologist A in order to achieve more accurate labels for the cytopathological sub-images.

[0069] This step is to reduce the number of false positive samples in object detection. Therefore, the CNN network is first used for preliminary screening. Since the input of the CNN network is generally small, it is considered to first input the patient WSI obtained in Step 1 into the openslide image segmentation tool for segmentation, and an image dataset of 1024×1024 pixels with different numbers is obtained for different patients. Since it is a common phenomenon that it is difficult to extract image features from medical image data, we use the VGG model pre-trained with the ImageNet image dataset as the backbone network, and add channel attention and spatial attention modules to both sides of the network. Finally, the feature maps of the three channels are fused as the extracted features, and then the high-dimensional features are extracted for k-means mean clustering. The clustering centers are used as the codebooks, and then the statistical vectors of the codebooks are used as the features of the samples. Finally, the samples are input into the SVM classifier for classification. The final model realizes the preliminary screening and classification of high-precision positive and negative samples.

[0070] For the selection of the clustering centers of K-means mean clustering here, we conducted experiments at equal intervals from 5 to 100 respectively. The experimental results show that the classification accuracy first increases and then decreases with the increase of the clustering centers. When the clustering centers are selected as 20, the accuracy performance is the best, which is 81.5%.

[0071] Step 3: Use the improved Faster R-CNN object detection model to perform intelligent detection of cancer cells.

[0072] Read all the sub-images in Folder 1 processed in Step 2 into the computer, and at the same time read the trained improved Faster R-CNN model to perform one recognition on the images. The samples used here are single-object recognition, only detecting positive targets. The network output is all the recognized sub-images in Folder 1, which are saved in Folder 3. The training process also requires two pathologists to label the samples in Folder 1. The specific labeling process is as follows:

[0073] ① Read all the sub-images into the random cropping program to make a unified dataset that meets the regulations

[0074] ② Open the sub-images with the label Image tool and frame the positive target areas considered by the physician

[0075] ③ Convert the labeled data into the COCO dataset for backup through the conversion program

[0076] The training process mainly includes the adjustment of hyperparameters to minimize the loss of the network. Some training tricks are used here, mainly including preventing overfitting and non-maximum suppression.

[0077] This step uses the data with low false positives processed in step 3 as the recognition sample. In this sample, it is found through statistics that the size and aspect ratio of the target are not conducive to the use of the target detection model, mainly manifested in:

[0078] ① The target scales of the data set vary greatly, with the maximum and minimum values differing by nearly dozens of times. The drastic changes in the target scales bring certain difficulties to detection.

[0079] ② Count the target aspect ratio. The target aspect ratio is mainly concentrated in the range of 0.5 to 2, but there are still a certain number of extreme targets. Extreme targets in both scale and aspect ratio will increase the difficulty of anchor design in the commonly used anchor-based model.

[0080] ③ The size of the target relative to the image area. Most targets are very small compared to the image area. They must be specially processed before they can be trained.

[0081] For the above problems in the sample, we use the online random clipping method to deal with them. The main steps are as follows:

[0082] ① Randomly select a target in the input image, randomly cut out a sub-image with a side length ranging from 256 to 512 around the target, then scale it to a side length of 512 before sending it to the network.

[0083] ② If the side length of the target exceeds the range, the target and a small amount of background will be directly cut out and then scaled.

[0084] The model selected in this step is the two-stage Faster R-CNN model. The Faster R-CNN model evolved from R-CNN and Fast R-CNN. Due to its variable network structure, it is widely used in target detection tasks. In order to adapt to the detection task of endometrial cancer cells, we have improved the model. The main improvements are as follows:

[0085] ① Use DCN deformable convolution to improve the detection ability of irregular abnormal cells.

[0086] ②Using Global Context to enhance global information helps to judge cell lesions.

[0087] ③ Use OHEM to enhance learning on difficult samples.

[0088] Step 4: Count various indicators, generate blocks from WSI, send them to the network for prediction, and then merge the recognition results

[0089] Load all the sub - images in folder 3 processed in step 3 and folder 1 in step 2 into the same folder, restore the size of the sub - images to the original size, and restore the original WSI in sequence according to the line numbers and column numbers included in the sub - image file names. The overall flowchart of the algorithm is as shown in Figure 6 shown below.

[0090] This step is to complete the last step of patient diagnosis. The method of the present invention is aimed at the WSI of patients, and it is also necessary to restore the experimental samples in the previous three steps. When using the openslide tool to cut the WSI in step 2, the name of each image follows the naming convention of patient number_line number_column number. Therefore, we restore the identified images processed in steps 2 and 3 to the size of 1024×1024. According to the line numbers and column numbers, the WSI can be completely restored. This WSI contains the positions and confidence levels of the cancer cells identified by the network, completing the intelligent detection of the endometrial cell pathological images of the patient.

[0091] The selected detection indicators mainly include:

[0092] ①Initial screening stage: accuracy, recall rate, precision, specificity

[0093] ②Recognition stage: AP, mAP

[0094] The method of the present invention detects cancer cells on a patient - by - patient basis, and the image analysis is based on the patient's WSI image. First, read all the images after cutting of the patient, and input the dataset after the initial screening by CNN into the object detection model to locate the positions of the cancer cells and give the confidence levels. Therefore, the main indicators for judging the effect of the method of the present invention are the sensitivity of the initial screening and the accuracy of the object detection. The mathematical formulas for the main indicators are as follows:

[0095] Initial screening stage:

[0096]

[0097]

[0098]

[0099]

[0100]

[0101] Recognition stage:

[0102] Based on the Precision - Recall curve, by calculating the average value of the Precision values corresponding to each recall value, an evaluation in the form of an array can be obtained, which is AP:

[0103]

[0104] where r1, r2, …, r n is the Recall value corresponding to the first interpolation point of the Precison interpolation segment arranged in ascending order. The AP for all categories is mAP. The mAP formula is:

[0105]

[0106] The computer device provided by an embodiment of the present invention. The computer device of this embodiment includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps in each of the above method embodiments. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in each of the above device embodiments.

[0107] The computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention.

[0108] The computer device can be a computing device such as a desktop computer, a notebook, a palm computer, and a cloud server. The computer device may include, but is not limited to, a processor and a memory.

[0109] The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0110] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the computer device by running or executing the computer program and / or module stored in the memory, and by calling the data stored in the memory.

[0111] If the modules / units integrated in the computer device are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, to implement all or part of the processes in the above-described embodiment methods of the present invention, it can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the above-described various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disc, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0112] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting endometrial cancer cells, characterized in that, Including the following steps: Using a CNN model to perform a preliminary screening on the WSI images of endometrial cell pathological specimens, and screening out the images containing positive cell samples; the CNN network is a CNN model based on Attention and Bag of words; the CNN model uses the VGG network as the backbone network, and channel attention and spatial attention modules are added to both sides of the backbone network respectively. Finally, the feature maps of the three channels are fused as the extracted features, and then the high-dimensional features are extracted for k-means mean clustering. The clustering centers are used as the codebook, and then the statistical vectors of the codebook are used as the features of the samples. Finally, they are input into the SVM classifier for classification, and the final model realizes the preliminary screening classification of positive and negative samples; Using the Faster R-CNN object detection model to identify the images of positive cell samples, specifically including processing by the method of online random cropping; randomly selecting a target in the input image of the positive cell sample, randomly cutting out a sub-image with a side length in the range of 256 to 512 pixels around the target, and then scaling it to a side length of 512 pixels and inputting it into the Faster R-CNN object detection model for identification; if the side length of the target exceeds 512 pixels, the target and the background are directly cut out and then scaled; the Faster R-CNN object detection model uses DCN deformable convolution to improve the detection ability for irregular abnormal cells; uses Global Context to enhance global information, which helps to judge cell lesions; uses OHEM to enhance the learning of difficult samples; Restoring the WSI image corresponding to the identified endometrial cancer cells to obtain the position and confidence of the endometrial cancer cells.

2. The endometrial cancer cell detection method according to claim 1, characterized in that, The WSI image is collected from the endometrial cell specimen by a Li-brush endometrial cell collector and scanned by a digital scanner to obtain the WSI image.

3. The endometrial cancer cell detection method according to claim 1, wherein The step of using a CNN model to perform a preliminary screening on the WSI images of endometrial cell pathological specimens and screening out the images containing positive cell samples includes: First, the WSI image is segmented using the openslide tool. Input a complete cell pathological WSI image, and output several 1024×1024 image patches; input the image patches into the CNN network for classification screening, and output the results of the preliminary screening.

4. The endometrial cancer cell detection method according to claim 1, characterized in that, The step of restoring the WSI image corresponding to the identified endometrial cancer cells to obtain the position and confidence of the endometrial cancer cells includes: Restoring the image of the positive cell sample to 1024×1024 pixels, and completely restoring the WSI image of the endometrial cell pathological specimen according to the row number and column number. This WSI image contains the position and confidence of the cancer cells identified by the Faster R-CNN object detection model.

5. An endometrial cancer cell detection system for implementing the method described in claim 1, characterized in that, Including: An image acquisition module for using a CNN model to perform a preliminary screening on the WSI images of endometrial cell pathological specimens and screening out the images containing positive cell samples; An identification module for using the Faster R-CNN object detection model to identify the images of positive cell samples; A position and confidence calculation module, which is used to restore the WSI image corresponding to the identified endometrial cancer cells to obtain the position and confidence of the endometrial cancer cells.

6. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1-4 are implemented.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1-4 are implemented.

Images